Devising a Diet Scoring Algorithm Sensitive to the Entire Range of Dietary Intakes

Abstract

ObjectivesMost, if not all, current diet scoring algorithms are not sensitive to the entire range of reported dietary intakes. That is, differences in intake do not always result in different scores. In some cases, the scoring function is binary: either a dietary intake met some criterion or it did not. Other scoring functions, such as the Healthy Eating Index (HEI), provide continuous scoring over a range of dietary intakes but become insensitive to changes at some high or low intakes. For example, according to estimates from the 2011–2012 National Health and Nutrition Examination Survey, about 8% of U.S. adults had diets that received the minimum score (0) for sodium. Information theory suggests there may be latent correlation or explanatory value in preserving the entire intake range in the scores. Therefore, the objective of this work was to devise a diet scoring algorithm sensitive to the entire range of dietary intakes. MethodsDesirable properties of a new scoring function were identified to be: (1) sensitive to differences in the entire range of dietary intakes (does not lose any information); (2) mathematically friendly with a continuous first derivative and unique inverse everywhere, thereby permitting statistical operations in both its domain and range; (3) plausibly related to biological processes; (4) not as dependent upon the accuracy and precision of the dietary standards employed; (5) simple to express and calculate; and (6) to otherwise mimic the scoring system of the Healthy Eating Index. ResultsThe proposed scoring function for all adequacy components, where greater intake is awarded higher scores, is 1-exp(-2x), where x is the normalized intake with 1.0 representing the desirable intake standard. The proposed scoring function for moderation components, where greater intake is awarded lower scores, uses a “most sensitive intake” parameter y and a scoring slope parameter k, set using professional judgment while considering standards and the range of intakes. For example, the moderation function for saturated fats uses y = 2, k = 2, and is expressed as 1–0.5 * exp(-k(y-x)) for x < y and 0.5 * exp(-k(x-y)) for x ≥ y. ConclusionsThe proposed scoring functions possess the six desired properties. They may serve to better track health-related outcomes over a greater range of intakes, and they will certainly be more statistically friendly. Funding SourcesNone.